基于加权主成分分析和高斯混合模型的神经元峰电位分类

Spike potential of nerve cell sorting based on weighted-PCA and Gaussian mixture model

通过加权主成分分析来对峰电位进行提取特征和降维,再利用高斯混合模型聚类算法对特征进行聚类,实现峰电位分类。采用开放的仿真数据和开放的实测数据分析验证算法的可行性和分类精度,并与主成分分析提取特征的高斯混合模型(GMM)聚类和加权主成分分析提取特征的K均值聚类2种方法进行了比较。仿真数据实验中,在噪声水平为0.05,0.10,0.15,0.20时,误分率分别为1.26%,1.43%,2.32%和3.37%,低于其他2种方法;实测数据实验中,恒河猴数据的平均J 3准则值为14.12,与其他2种方法相比,平均J 3准则值较大。

The spike signal is extracted by weighted principal component analysis to extract features and to reduce the dimension.The features are clustered by Gaussian mixture model clustering algorithm.Spike sorting is achieved.Open simulation data and open measured data are used to analyze and verify the classification precision and feasibility of the algorithm and compared with the two methods of the Gaussion mixture model(GMM)clustering of the principal component analysis(PCA)and the K-means clustering of weighted PCA.In simulation data experiment,when the noise levels are 0.05,0.10,0.15,and 0.20,the misclassification rates are 1.26%,1.43%,2.32%,and 3.37%,respectively,which are lower than the other two methods.In the measured data experiment,the average J 3 criterion value of the Rhesus monkey data is 14.12.Compared with the other two methods,the average J 3 criterion value is larger.